p.2 #1 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Excellent idea to capture the behavior on video.
I have an R6 rather than an R5, but I've often used my dogs as test subjects for animal eye tracking. Most of the time the tracking indicator and the actual focus stay in close sync but I **have** - very occasionally - experienced something like the behavior shown in the video around the 40 sec mark.
To be clear my experiences haven't been as pronounced and long-lived as the example in the video, but that may be because I instinctively disengage then re-engage the AF whenever I see any kind of AF misbehavior (long standing BBF user). To be honest I tend to "pump" the AF periodically even if the tracking lock seems to be holding; it's a voodoo-like habit I acquired well before mirrorless came along and I haven't ditched it yet.
Regardless, I do believe you've caught a quirk in the AF algorithms and the best chance of getting it fixed is to share the video with Canon - so please do that!
p.2 #2 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
The AF assist lamp is on, not that it matters as it would never turn on in this circumstance, and the lens and camera were perfectly capable of acquiring focus on the eye and the nose repeatedly.
To follow up on why the video was dark, I had accidentally left the CLog Gamma and BT2020 settings used for recording video on, and the output for the viewfinder is standard rec709. Whoops.
I took a bunch of videos today testing out various lenses. It frequently jumps to focus on the nose using any lens I tested with (RF 35mm f/1.8, RF 24-70 f/2.8L, RF 70-200 f/2.8L). However, it doesn't get stuck the way that the RF 50mm f/1.2L.
The RF 50mm f/1.2L exhibited the same behavior in broad daylight while doing these tests today.
Here an example of it happening today while we were out. The image was clearly focused on the eye, but the moment before I pressed the shutter button it refocused to the nose and stayed till I released the AF button and reacquired focus. At least for my dog, something very odd is going on.
p.2 #3 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
The nose looks like a little face with nostrils for eyes?
Actually though if you think about this, Canon started with people face detect before eye AF. Perhaps it is reverting to ITR mode and face detect finding the nose and it thinks eyes should be really close to the nose, for people. My theory anyways, but it doesn't explain the lens being part of the equation.
p.2 #6 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
TeamSpeed wrote:
The nose looks like a little face with nostrils for eyes?
I have noticed with my own dogs (Beagles) that if the nose has a better / more pronounced catchlight than either of the eyes then the camera will be drawn to it, especially in low / subdued light. Typically the AF box will show this happening; it's much less common for the AF indicator to be sticking to the eye while the nose is (very obviously) in focus - but it has happened.
p.2 #7 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
In some of the early videos showing eye AF from reviewers that as dogs would run away from there camera, eye AF detected their rear as an eye. It was pretty funny.
p.2 #8 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
TeamSpeed wrote:
In some of the early videos showing eye AF from reviewers that as dogs would run away from there camera, eye AF detected their rear as an eye. It was pretty funny.
That video illustrated the "brown eye" deep leaning algorithm.
That was indeed funny as hell.
p.2 #10 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Hopefully, the op is developing a clearer understanding of the af capabilities and limitations of this camera. One additional suggestion is to modify the case parameters tracking sensitivity and/or accel - decel. Since the af is jumping to the nose, I suggest turning it down a notch to -1 or even -2. This will make the af more “sticky” and decrease the tendency to jump to the nose. As we’ve discussed, the AED is an algorithmic based method of selecting the eye based on shape, color, and other parameters. Further as we’ve stated, it has a tendency to struggle more on dark eyes (try a bison for example - brown on brown). A pink circular nose in the front of the face of an animal with dark eyes is a likely target. Pumping the af to reacquire is another strategy as is using spot or zone. FWIW, my Sony gear exhibited similar difficulties and even the a1 from what I’ve seen has similar limitations.
p.2 #12 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Jesse Evans wrote:
Here an example of it happening today while we were out. The image was clearly focused on the eye, but the moment before I pressed the shutter button it refocused to the nose and stayed till I released the AF button and reacquired focus. At least for my dog, something very odd is going on.
So let us know what Canon tells you. If this is a known failure point, they will tell you to just send in the camera. If it is uncommon, most likely they will tell you to reset the camera and see what happens.
Did you reset the camera? Did you call Canon?
I love posting my issues here as I learn alot... but before I get all twisted up in my shorts... I ALWAYS call Canon either before I post here or shortly thereafter. Unless someone posts something like "set menu X item to Y" to make the issue go away, I call Canon too.
Let us know. As I noted before, I think your camera is broken. While all this expert opinion is fascinating, I would still call Canon and after going though more than a few calls with them, I save some time by just resetting my camera first....because they always told me to do that. YMMV.
p.2 #13 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Even if it is ultimately determined this is just the limit of the AF system with all the variables it must account for, Canon still needs to change one thing.
If the PDAF pixels on the nose are being used for focus, then move the displayed AF box so the user knows where focus is occurring.
Showing the user "Here are the eyes!" while deep in the AF algorithm deciding "I'll pick the nose!" isn't very user-friendly.
p.2 #14 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
I've recently been having a bit of a problem getting "keepers" while shooting photos of my fast-moving Brittany pup in bright, albeit harsh light. I think I know why, but I'm having fun doing experiments. We know about low-light problems for AF, but I'm wondering if very bright light combined with bright white snow and a little pup coming almost straight at me is a perfect storm for erratic AF. I keep my shutter speed up quite high, at least 1/4000th. It appears the AF point is on or very near the eye, but many images are a tad bit soft.
Lots to experiment with on a non-wild subject that loves to play and run. But I think the fast-moving tiny pup eyes in really bright conditions might also be a bit too much for this computer of a camera to handle at times.
p.2 #15 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Jesse Evans wrote:
The AF assist lamp is on, not that it matters as it would never turn on in this circumstance, and the lens and camera were perfectly capable of acquiring focus on the eye and the nose repeatedly.
To follow up on why the video was dark, I had accidentally left the CLog Gamma and BT2020 settings used for recording video on, and the output for the viewfinder is standard rec709. Whoops.
I took a bunch of videos today testing out various lenses. It frequently jumps to focus on the nose using any lens I tested with (RF 35mm f/1.8, RF 24-70 f/2.8L, RF 70-200 f/2.8L). However, it doesn't get stuck the way that the RF 50mm f/1.2L.
The RF 50mm f/1.2L exhibited the same behavior in broad daylight while doing these tests today.
Here an example of it happening today while we were out. The image was clearly focused on the eye, but the moment before I pressed the shutter button it refocused to the nose and stayed till I released the AF button and reacquired focus. At least for my dog, something very odd is going on....Show more →
Coming to this late. My first reaction after watching the video is the PDAF was losing a lock on the eye. Eye detection occurs in the spatial domain, which is separate from PDAF's ability to acquire. But then I read your post about this occurring in daylight, so there shouldn't be any reason for the PDAF to fail to acquire...unless there is insufficient vertical contrast (irrespective of light levels), which is typically the case for dark eyes but I imagine Canon's algorithm is smart enough to use areas around the eye instead.
Q: Does Canon's eye-detect AF implementation reflect the state of PDAF acquisition in the color of the box around the eye, ie change from one color to another when PDAF fails or succeeds? I'm guessing no from the video but just wanted to make sure since that fact would be useful in determining if this is a PDAF acquisition issue.
p.2 #16 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
snapsy wrote:
Coming to this late. My first reaction after watching the video is the PDAF was losing a lock on the eye. Eye detection occurs in the spatial domain, which is separate from PDAF's ability to acquire. But then I read your post about this occurring in daylight, so there shouldn't be any reason for the PDAF to fail to acquire...unless there is insufficient vertical contrast (irrespective of light levels), which is typically the case for dark eyes but I imagine Canon's algorithm is smart enough to use areas around the eye instead.
Q: Does Canon's eye-detect AF implementation reflect the state of PDAF acquisition in the color of the box around the eye, ie change from one color to another when PDAF fails or succeeds? I'm guessing no from the video but just wanted to make sure since that fact would be useful in determining if this is a PDAF acquisition issue....Show more →
If focus acquisition failed the box turns red.
White: not focusing
Blue: successfully focusing
Red: failing to acquire focus
p.2 #17 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Jesse Evans wrote:
If focus acquisition failed the box turns red.
White: not focusing
Blue: successfully focusing
Red: failing to acquire focus
Thanks. So based on what you're observing, the camera reliably detects the eye, indicates it has successfully focused on the eye, yet sometimes switches focus to the nose instead. Btw, how quickly does the camera respond to changes in its focus success (ie, change in color) in your experience?
p.2 #19 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
Bruce n Philly wrote:
So let us know what Canon tells you. If this is a known failure point, they will tell you to just send in the camera. If it is uncommon, most likely they will tell you to reset the camera and see what happens.
Did you reset the camera? Did you call Canon?
I love posting my issues here as I learn alot... but before I get all twisted up in my shorts... I ALWAYS call Canon either before I post here or shortly thereafter. Unless someone posts something like "set menu X item to Y" to make the issue go away, I call Canon too.
Let us know. As I noted before, I think your camera is broken. While all this expert opinion is fascinating, I would still call Canon and after going though more than a few calls with them, I save some time by just resetting my camera first....because they always told me to do that. YMMV.
Nobody is getting all twisted up, it is just a discussion thread. Much like yours when you initially had eye AF/AI Servo issues when you first got your mirrorless, and you asked for help here to point you to settings or suggestions to try. Did you call Canon immediately, or did you throw it out to the forum?
The camera isn't broken, it just has alot of software behind the scenes along with a huge machine learning matrix to dig through that it was loaded up with that is likely causing this. Future firmware updates can (and will) change much of this as Canon changes the deep learning content that was initially loaded upon the cameras' release.
Also whomever you call at Canon and get on the phone are likey just tier 1 support (if that) and they are just going to follow a script, they will likely be less of an expert than some here on the boards. I have lost track of the times somebody at Canon told someone a "fact" or suggested "a fix", only to have others prove that wrong, or it was obviously the wrong answer.
p.2 #20 · Canon R5 - Tracking Box is Inconsistent with Results - Animal Eye AF
I called Canon. The next day after posting. I wanted to know, from the forum if this camera was flaky.. I was having some buyers remorse.
It was my 4th Canon body and I never wrestled this much with a body before.
Peace,
Bruce in Philly
TeamSpeed wrote:
Nobody is getting all twisted up, it is just a discussion thread. Much like yours when you initially had eye AF/AI Servo issues when you first got your mirrorless, and you asked for help here to point you to settings or suggestions to try. Did you call Canon immediately, or did you throw it out to the forum?
The camera isn't broken, it just has alot of software behind the scenes along with a huge machine learning matrix to dig through that it was loaded up with that is likely causing this. Future firmware updates can (and will) change much of this as Canon changes the deep learning content that was initially loaded upon the cameras' release.
Also whomever you call at Canon and get on the phone are likey just tier 1 support (if that) and they are just going to follow a script, they will likely be less of an expert than some here on the boards. I have lost track of the times somebody at Canon told someone a "fact" or suggested "a fix", only to have others prove that wrong, or it was obviously the wrong answer. ...Show more →
